Conversation
When both trace(g) and trace(g^-1) were nonzero the function returned early without testing the coefficient identities, so it never ruled out a form. Over a large field almost every element has nonzero trace, which left recog's classical recognition of SL(6,257) undecided about half the time (recog issue #492). Nonzero traces determine the scalar lambda relating g^-1 to g directly, so use it in the existing coefficient test instead of returning. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## master #88 +/- ##
==========================================
+ Coverage 80.40% 81.79% +1.38%
==========================================
Files 8 8
Lines 3670 3669 -1
==========================================
+ Hits 2951 3001 +50
+ Misses 719 668 -51
🚀 New features to boost your workflow:
|
| if not IsZero(tM) and not IsZero(tMi) then | ||
| return [i0, tM/tMi]; | ||
| fi; |
There was a problem hiding this comment.
This is the most important bit which previously prevented us from ruling out forms in many cases.
|
@jdebeule I'd like to merge this and make a release this week so we can get this into GAP 4.17.0, is that OK? I discussed it thoroughly with @aniemeyer (who is sitting next to me right now), and this function is not used by forms itself, only by recog; and is not even documented in the forms manual. The fix is important because before we failed to rule out the presence of non-trivial forms for e.g. SL(6,257) about 50% of the time. I think future recog versions will not use this anymore, it's easier to just copy this code to recog; but for the sake of already released recog versions, it'd be good to fix it here (and leave it in here for the foreseeable future, the copy in |
This branch has not been deployed
1 participant
When both trace(g) and trace(g^-1) were nonzero the function returned early without testing the coefficient identities, so it never ruled out a form. Over a large field almost every element has nonzero trace, which left recog's classical recognition of SL(6,257) undecided about half the time (see gap-packages/recog#492).
Nonzero traces determine the scalar lambda relating g^-1 to g directly, so use it in the existing coefficient test instead of returning.
Co-Authored-By: Claude Fable 5.1 noreply@anthropic.com